In the transport market, established networks are often organized into ring topologies based upon SONET. Although suffering the limitations of a circuit-oriented system, SONET provides important management and reliability capabilities that make the provisioning of a nationwide network comprising thousands of nodes feasible and economic. The equipment that bridges traffic on and off these rings is the Add-Drop Mux (ADM). Traditionally these supplied SONET connections both along the ring and from the feeder ports. The SONET circuit conventions were preserved throughout the system, so that both the switch ports and the fabric interconnecting them were circuit-oriented. This made particular sense when voice circuits comprised most of the feeder traffic.

But as packet data has grown to dominate the feeder traffic, there is economy in pushing the packet-awareness into the ADM and isolating SONET to the transport domain. To enable an easy migration, vendors developed the Multi-Service Provisioning Platform (MSPP), a converged system that employs parallel switching fabrics and interface logic, one for circuits on the ring and the other for packet connections among the feeder ports. By integrating Layer 2 packet technology into the MSPP, fewer expensive router ports are required to support connectivity to the transport domain.

One step beyond the dual fabric is a redesign of the transport network itself, architecting it around packet transport while creating structures to support legacy circuits; this is the promise of RPR. In a nutshell, RPR melds the resiliency and management advantages of SONET onto a packet ring network. Designed as a packet-based alternative to SONET, RPR is gaining particular momentum as an attractive aggregation mechanism for residential and business access.

Using a packet ring to aggregate packet data is highly efficient, because the bursty nature of packet data allows extensive statistical multiplexing of the ring capacity. Instead of dedicating a fixed circuit to a particular customer, the carrier can oversubscribe the ring bandwidth among a large population of users with reasonable certainly that the traffic patterns won’t overlap. For traffic sensitive to packet loss, such as voice or network control, RPR provides several configurable levels of QoS that convey strict guarantees of an appropriate amount of ring bandwidth.

Here is another area where integrated silicon solutions can assist an evolutionary convergence of the network. RPR interfaces can be easily combined with both the packet and circuit switch fabrics in existing ADM's, such that these devices can be upgraded to support RPR rings in parallel with their existing connections or to replace them. The carrier can transition to an RPR access or metro strategy without having to risk an expensive and tricky wholesale upgrade.

As these examples show, QoS is one of the keys to successful convergence. Although circuit networks are sub-optimal for carrying packet data, they provide extremely reliable and predictable service quality. Once a customer has paid for a circuit and the provider has bolted it down through their network, the user will always get exactly that amount of bandwidth whether they use it or not. This guarantee is exceptionally valuable, particularly for voice transport, and the challenge for packet networks is to match this capability.

With the transition to VOIP in the enterprise, this service guarantee reaches center stage. Most businesses rely on their phone network 24/7, and it can be financially painful and in some cases a breach of contract for those phone connections to go out of service. However, if voice traffic is blindly merged onto the existing enterprise LAN, service interruptions are likely and will occur with greatest frequency during the working day when the network is most utilized and the phone connections most critical.

Ethernet is an inherently bursty protocol; the combination of many users sharing random access to the network produces spikes in utilization. Another important, and often overlooked, aspect of Ethernet is the absence of an access reservatio--a node on the network may begin transmitting whenever it chooses, and for however long. Because most LANs are organized in a hub and spoke system, the total interconnect bandwidth is far less than the sum of all the bandwidth on each user's desktop. If all users chose to send the full bandwidth at the same time, the core of the network lacks the capacity to complete the transmissions and will drop packets; this is where quality of service enters the picture.

This oversubscription of the core switching bandwidth need not degrade application utility. Because of the retransmission policies of the TCP protocol, most services are resilient to packet loss and don’t require reserved bandwidth on the network. It is only real-time communications, such as voice or videoconferencing and network control traffic, that relies on low latency and minimal packet drops. By carefully classifying the traffic stream, and distinguishing the critical packets, the networking equipment can guarantee that only the traffic that truly needs preferential access to the network receives it.

Line card count and productivity
To service both the old world and the new, potentially each port configuration and each protocol requires a new line card. Yet, developing a card for each point solution is enormously inefficient, and ignores the extensive commonality in the underlying technology of each card. A better solution is possible: architect both the system and the underlying ICs to leverage common technology for multiple end products.

A new line card development is an enormous expenditure of R&D dollars and human resources. After IC design, board design, NRE charges, board/IC spins, and qualification cycles an equipment vendor can easily spend in excess of $10 million developing a new line card. But the costs don't end there: once that line card is released into production it must be separately procured, tested, tracked, and inventoried. After the card leaves the manufacturer and enters the distribution system, the tracking and inventory costs are incurred once again. Even in the hands of the end customer, multiple discrete cards can create a qualification and inventory nightmare.

Two factors can dramatically reduce the impact of this: an emphasis on convergence and the emergence of multi-protocol silicon. If the system vendor understands this dynamic, they can architect their cards with an eye towards identifying the common functions inherent in each. And the component vendor can dramatically influence the costs incurred in the system by tailoring their IC design to suit this multi-protocol, multi-rate environment. The situations described above illustrate how the IC and system industries can work together to facilitate convergence efficiently and with minimal disruption to the installed base.

Conclusions
As the communications industry transitions from a multi-protocol, multi-network system to a converged, packet-based, unified system, communications equipment will need to bridge the gap between the old and the new. Highly integrated networking silicon provides a means to cost-effectively service different protocols from a single IC, reducing the quantity of line card developments and improving vendor return on investment. Advanced architectural techniques, such as deep packet inspection and traffic management, provide the necessary quality of service for packet-based networks to achieve the converged vision and replicate the advantages of circuit switching with packet technology.

For equipment vendors, revenue growth in the latter part of the decade will depend upon a ruthless focus on productivity improvement. The true test of any technology is whether it solves important problems more cheaply and effectively than current means. For equipment vendors to increase the size of the networking pie they must embrace a converged vision that enables a cost-effective evolution from the multi-protocol world of the past to the unified network of the future.

About the Author
Jim McKeon a product manager for Ethernet devices at Cortina Systems. He holds A.B. and B.E. degrees from Dartmouth College and an MSEE from the University of Texas at Austin. Jim can be reached at: jim@cortina-systems.com